• Share
  • Email
  • Embed
  • Like
  • Save
  • Private Content
Monetizing User Activity on Social Networks
 

Monetizing User Activity on Social Networks

on

  • 2,480 views

Monetizing User Activity on Social Networks - Challenges and Experiences, 2009 IEEE/WIC/ACM International Conference on Web Intelligence, Sep 15-18 2009

Monetizing User Activity on Social Networks - Challenges and Experiences, 2009 IEEE/WIC/ACM International Conference on Web Intelligence, Sep 15-18 2009

Statistics

Views

Total Views
2,480
Views on SlideShare
2,299
Embed Views
181

Actions

Likes
2
Downloads
37
Comments
0

5 Embeds 181

http://knoesis.wright.edu 169
http://www.slideshare.net 5
https://www.linkedin.com 3
http://www.lmodules.com 2
http://www.linkedin.com 2

Accessibility

Upload Details

Uploaded via as Adobe PDF

Usage Rights

© All Rights Reserved

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Processing…
Post Comment
Edit your comment

    Monetizing User Activity on Social Networks Monetizing User Activity on Social Networks Presentation Transcript

    • Meena  Nagarajan,  Amit  P.  Sheth                                                                    KNO.E.SIS  Center                                                Wright  State  University   M.  Nagarajan,  K.  Baid,  A.  P.  Sheth,  and  S.  Wang,  "Monetizing  User  Activity  on  Social  Networks  -­‐   Challenges  and  Experiences“,  2009  IEEE/WIC/ACM  International  Conference  on  Web   Intelligence,  Milan,  Italy  
    •   On  social  networks     Use  case  for  this  talk       Targeted  content  =  content-­‐based  advertisements       Target  =  user  profiles     Content-­‐based  advertisements  CBAs     Well-­‐known  monetization  model  for  online   content  
    • May  30,June  02  2009  
    • June  01,  2009  
    •   Interests  do  not  translate  to  purchase  intents     Interests  are  often  outdated..     Intents  are  rarely  stated  on  a  profile..       Cases  that  work     New  store  openings,  sales     Highly  demographic-­‐targeted  ads    
    • June  01,  2009  
    • CONTENT-­‐BASED  ADS   ON  THEIR  PROFILES   June  01,  2009  
    •   Non-­‐trivial     Non-­‐policed  content   ▪  Brand  image,  Unfavorable  sentiments1     People  are  there  to  network   ▪  User  attention  to  ads  is  not  guaranteed     Informal,  casual  nature  of  content   ▪  People  are  sharing  experiences  and  events   ▪  Main  message  overloaded  with  off  topic  content   I  NEED  HELP  WITH  SONY  VEGAS  PRO  8!!  Ugh  and  i  have  a  video  project  due  tomorrow  for  merrill  lynch  :((  all  i   need  to  do  is  simple:  Extract  several  scenes  from  a  clip,  insert  captions,  transitions  and  thats  it.  really.  omgg  i  cant   figure  out  anything!!  help!!  and  i  got  food  poisoning  from  eggs.  its  not  fun.  Pleasssse,  help?  :(   1Learning  from  Multi-­‐topic  Web  Documents  for  Contextual  Advertisement,  Zhang,  Y.,  Surendran,  A.  C.,  Platt,  J.  C.,  and  Narasimhan,  M.    ,  KDD  2008    
    •   Cultural  Entities   HOW     Word  Usages  in  self-­‐ presentation   WHY     Slang  sentiments   WHAT     Intentions  
    •   Identifying  intents  behind  user  posts  on  social   networks     Content  with  monetization  potential     Identifying  keywords  for  advertizing  in  user-­‐ generated  content     Interpersonal  communication  &  off-­‐topic  chatter  
    •   User  studies     Hard  to  compare  activity  based  ads  to  s.o.t.a     Impressions  to  Clickthroughs     How  well  are  we  able  to  identify  monetizable  posts     How  targeted  are  ads  generated  using  our  keywords   vs.  entire  user  generated  content  
    • Identification,  Evaluation  
    •   Scribe  Intent  not  same  as  Web  Search  Intent1     People  write  sentences,  not  keywords  or  phrases     Presence  of  a  keyword  does  not  imply   navigational  /  transactional  intents     ‘am  thinking  of  getting  X’  (transactional)     ‘i  like  my  new  X’  (information  sharing)     ‘what  do  you  think  about  X’  (information  seeking)   1B.  J.  Jansen,  D.  L.  Booth,  and  A.  Spink,  “Determining  the  informational,  navigational,  and  transactional  intent  of  web  queries,”  Inf.  Process.   Manage.,  vol.  44,  no.  3,  2008.  
    •   Action  patterns  surrounding  an  entity     How  questions  are  asked  and  not  topic  words   that  indicate  what  the  question  is  about     “where  can  I  find  a  chotto  psp  cam”     User  post  also  has  an  entity  
    • Set  of  user  posts  from  SNSs   Not  annotated  for  presence  or  absence   of  any  intent  
    • Generate  a  universal  set  of  n-­‐gram   patterns;  freq  >  f      S  =  set  of  all  4-­‐grams;  freq  >  3  
    • Generate  set  of  candidate  patterns  from   seed  words      (why,when,where,how,what)   Sc  =  all  4-­‐grams  in  S  that  extract  seed   words  
    • User  picks  10  seed  patterns  from  Sc     Sis  =  ‘does  anyone  know  how’,  ‘where  do  i   find’,  ‘someone  tell  me  where’….  
    • Sc  =  all  4-­‐grams  in  S  that   Sis  =  ‘does  anyone  know  how’,  ‘where   extract  seed  words   do  i  find’,  ‘someone  tell  me  where’….   Gradually  expand  Sis  by  adding   Information  Seeking  patterns  from  Sc  
    • Sis  =  ‘does  anyone  know  how’,  ‘where   do  i  find’,  ‘someone  tell  me  where’….   For  every  pis  in  Sis  generate  set  of  filler   patterns  
    • ‘.*  anyone  know  how’   ‘does  anyone  .*  how’   ‘does  .*  know  how’    ‘does  anyone  know  .*’   ‘does  anyone  know  how’   Look  for  patterns  in  Sc   -­‐ Functional  compatibility  of  filler   -­‐ words  used  in  similar  semantic  contexts   -­‐  Empirical  support  for  filler  
    •   Functional  properties  /  communicative  functions   of  words     From  a  subset  of  LIWC1     cognitive  mechanical  (e.g.,  if,  whether,  wondering,   find)     ▪  ‘I  am  thinking  about  getting  X’       adverbs  (e.g.,  how,  somehow,  where)       impersonal  pronouns  (e.g.,  someone,  anybody,   whichever)   ▪  ‘Someone  tell  me  where  can  I  find  X’     1Linguistic  Inquiry  Word  Count,LIWC,  http://liwc.net  
    •   Sc  =  {‘does  anyone  know  how’,  ‘where  do  I  find’,  ‘someone  tell  me   where’}     pis  =  `does  anyone  know  how’     ‘does  *  know  how’     ‘does  someone  know  how’   ▪  Functional  Compatibility  -­‐  Impersonal  pronouns   ▪  Empirical  Support  –  1/3     ‘does  somebody  know  how’   ▪  Functional  Compatibility  -­‐  Impersonal  pronouns   ▪  Empirical  Support  –  0   ▪  Pattern  Retained     ‘does  john  know  how’   ▪  Pattern  discarded  
    •   Over  iterations,  single-­‐word  substitutions,   functional  usage  and  empirical  support   conservatively  expands  Sis       Infusing  new  patterns  and  seed  words     Stopping  conditions  
    •   does anyone know how   no idea how to   anyone know how to   someone tell me how   i dont know what   have no clue what   know where i can   does anyone know if   tell me how to   i dont know if   i dont know how   know if i can   anyone know where i   anyone know if i   does anyone know where   im not sure if   does anyone know what   i was wondering if   anybody know how to   idea what you are   anyone know how i   let me know how   im not sure what   and i dont know   does anybody know how   now i dont know   does anyone know why   but i dont really   i was wondering how   was wondering if someone   does anyone know when   would like to see   tell me what to   see what i can   im not sure how   anyone have any idea   i was wondering what   wondering if someone could   was wondering how i   i do not want
    •   Information  Seeking  patterns  generated   offline     Information  seeking  intent  score  of  a  post     Extract  and  compare  patterns  in  posts  with   extracted  patterns     Transactional  intent  score  of  a  post   ▪  LIWC  ‘Money’  dictionary     ▪  173  words  and  word  forms  indicative  of  transactions,  e.g.,  trade,   deal,  buy,  sell,  worth,  price  etc.  
    •   Training  corpus     8000  user  posts   ▪  MySpace  Computers,  Electronics,  Gadgets  forum     309  unique  new  patterns,  263  unambiguous     Testing  patterns  for  recall     ‘To  buy’  Marketplace  –  average  81  %    
    • Off-­‐topic  noise  elimination  
    •   Identifying  keywords  in  monetizable  posts     Plethora  of  work  in  this  space     Off-­‐topic  noise  removal  is  our  focus   I  NEED  HELP  WITH  SONY  VEGAS  PRO  8!!  Ugh  and  i  have  a  video  project  due  tomorrow  for  merrill  lynch  :((  all  i   need  to  do  is  simple:  Extract  several  scenes  from  a  clip,  insert  captions,  transitions  and  thats  it.  really.  omgg  i  cant   figure  out  anything!!  help!!  and  i  got  food  poisoning  from  eggs.  its  not  fun.  Pleasssse,  help?  :(  
    •   Topical  hints     C1  -­‐  ['camcorder']     Keywords  in  post     C2  -­‐  ['electronics  forum',  'hd',  'camcorder',  'somethin',   'ive',  'canon',  'little  camera',  'canon  hv20',  'cameras',   'offtopic']     Move  strongly  related  keywords  from  C2  to  C1   one-­‐by-­‐one     Relatedness  determined  using  information  gain     Using  the  Web  as  a  corpus,  domain  independent  
    •   C1  -­‐  ['camcorder']     C2  -­‐  ['electronics  forum',  'hd',  'camcorder',   'somethin',  'ive',  'canon',  'little  camera',   'canon  hv20',  'cameras',  'offtopic']       Informative  words     ['camcorder',  'canon  hv20',  'little  camera',  'hd',  'cameras',   'canon']  
    • Preliminary  work  
    •   Keywords  from  60  monetizable  user  posts     Monetizable  intent,  at  least  3  keywords  in  content     45  MySpace  Forums,  15  Facebook  Marketplace,   30  graduate  students     10  sets  of  6  posts  each     Each  set  evaluated  by  3  randomly  selected  users     Monetizable  intents?     All  60  posts  voted  as  unambiguously  information   seeking  in  intent  
    •   Google  AdSense  ads  for  user  post  vs.   extracted  topical  keywords  
    •   Choose  relevant  Ad  Impressions     VW  6  disc  CD  changer         I  need  one  thats  compatible  with  a  2000  golf   most  are  sold  from  years  1998-­‐2004if  anyone   has  one  [or  can  get  one]  PLEASE  let  me  know!  
    •   Users  picked  ads  relevant  to  the  post     At  least  50%  inter-­‐evaluator  agreement     For  the  60  posts     Total  of  144  ad  impressions     17%  of  ads  picked  as  relevant     For  the  topical  keywords     Total  of  162  ad  impressions     40%  of  ads  picked  as  relevant  
    •   User’s  profile  information     Interests,  hobbies,  tv  shows..     Non-­‐demographic  information     Submit  a  post     Looking  to  buy  and  why  (induced  noise)     Ads  that  generate  interest,  captured   attention  
    •   Using  profile  ads     Total  of  56  ad  impressions     7%  of  ads  generated  interest     Using  authored  posts     Total  of  56  ad  impressions     43%  of  ads  generated  interest     Using  topical  keywords  from  authored  posts     Total  of  59  ad  impressions     59%  of  ads  generated  interest  
    •   User  studies  small  and  preliminary,  clearly   suggest     Monetization  potential  in  user  activity     Improvement  for  Ad  programs  in  terms  of  relevant   impressions     Evaluations  based  on  forum,  marketplace     Verbose  content     Status  updates,  notes,  community  and  event   memberships…     One  size  may  not  fit  all  
    •   A  world  between  relevant  impressions  and   clickthroughs     Objectionable  content,  vocabulary  impedance,  Ad   placement,  network  behavior     In  a  pipeline  of  other  community  efforts     No  profile  information  taken  into  account     Cannot  custom  send  information  to  Google   AdSense  
    •   Keywords  to  Ad  Impressions     Google  Adsense  like  web  service     Monetization  potential  of  a  keyword  on  the   Web  not  the  same  on  a  social  n/w?     Ranking  keywords  in  user  post     We  are  building  an  F8  app     Collaboration  for  clickthrough  data  
    •   Google/Bing:  Meena  Nagarajan     meena@knoesis.org     http://knoesis.wright.edu/students/meena/     Google/Bing:  Amit  Sheth     amit@knoesis.org     http://knoesis.wright.edu/amit